A deep intelligent framework for software risk prediction using improved firefly optimization

For the success of a software project, early and precise discrimination of software requirement risks are essential. Researchers have suggested several predictive methods based on conventional deep learning techniques, but the common factor is the high misclassification rate. Setting the appropriate hyperparameters is a complex problem in deep learning. Manual, grid and random searches are the most common methods for locating the best deep neural network hyperparameters. However, these are not the choice when experience is lacking. In this article, we proposed Deep Neural Network with Memetic firefly. The deep neural network performs classification tasks, and deep neural network hyperparameters have been optimized using a memetic firefly algorithm. Moreover, a novel perturbation operator is introduced to get rid of locally optimal solutions of the traditional firefly algorithm. The proposed method’s performance has been validated by comparing various hybrid techniques such as a deep neural network with firefly hill-climbing, a deep neural network with firefly, a deep neural network with PSO, and a deep neural network to identify the risk in software requirements. The proposed deep neural network with the memetic firefly algorithm outperforms all compared approaches with a 98.8% accuracy. It is adaptable for accurate software risk prediction in projects.